cells

Data and its logic, in one plain-text file.

A .cell file is a table that says what its columns are, computes what it can, and knows the files it points at. Plain text, checked by every reader.

1---
2accounts = ```cells
3account:text!
4bank
5rent
6food
7```
8---
9date:date,amount:EUR>0,from:accounts.account,to:accounts.account,vat:EUR=amount * 21 / 121
102026-11-01,950.00,bank,rent,
112026-11-03,42.50,bank,food,
12
NORmoney.cell1 sel 1:1
 

money.cell · Cells

	date	amount	from	to	vat
1	2026-11-01	950.00	bank	rent	164.88
2	2026-11-03	42.50	bank	food	7.38

$ cells check money.cell
money.cell row 2, to: "coffee" is no key of accounts
$ cells check money.cell && echo ok
ok
$ █
$ 

Cells00:00 PM

battery 82%·still awake

THE FILE ON THAT SCREEN

This is a cell. Everything it does is in its text.

money.cell

---
accounts = ```cells
account:text!
bank
rent
food
coffee
```
---
date:date,amount:EUR>0,from:accounts.account,to:accounts.account,vat:EUR=amount * 21 / 121
2026-11-01,950.00,bank,rent,
2026-11-03,42.50,bank,food,
2026-11-05,3.80,bank,coffee,

A CSV knows nothing. Every value is a string that hopes.

96.7x reads as fine, because nothing in the file says what cost is.

groceries.csv

item,cost,qty,paid
fuel,84.20,1,yes
groceries,96.7x,2,no
fuel,12.00,0,maybe

what Cells reads

	item	cost	qty	paid
1	fuel	84.20	1	yes
2	groceries	96.7x	2	no
3	fuel	12.00	0	maybe

every cell holds its type

Say what each column is. In the header, where it cannot drift.

The schema is the first line of the file, so every reader checks the same rules and names what breaks them.

try Fix the cells: 96.75, 1, no.

groceries.cell

item:text!,cost:EUR>=0,qty:int>=1,paid:yes/no
fuel,84.20,1,yesfuel"fuel" is the key of row 2 too
groceries,96.7x,2,no96.7x"96.7x" is not a number
fuel,12.00,0,maybefuel0maybe"fuel" is the key of row 0 too · "0" is below 1 · "maybe" is not yes or no

what Cells reads

	item	cost	qty	paid
1	fuel	84.20	1	yes
2	groceries	96.7x	2	no
3	fuel	12.00	0	maybe

row 0, item: "fuel" is the key of row 2 too
row 1, cost: "96.7x" is not a number
row 2, item: "fuel" is the key of row 0 too
row 2, qty: "0" is below 1
row 2, paid: "maybe" is not yes or no

Columns that compute. The rule beside the column it fills.

The field stays empty and the header holds the formula, so a value is never stored twice and never stale.

try Change a cost and watch total follow.

groceries.cell

item:text!,cost:EUR>=0,qty:int>=1,total:EUR=cost * qty
fuel,84.20,1,
groceries,96.75,2,
coffee,2.10,3,

what Cells reads

	item	cost	qty	total
1	fuel	84.20	1	84.20
2	groceries	96.75	2	193.50
3	coffee	2.10	3	6.30

every cell holds its type

A cell can hold its own. By name, never by coordinate.

There is no B7 to point at the wrong row after a line is inserted by hand: a formula names what it reads.

try Write `=96.75 / 4` instead.

groceries.cell

item:text!,cost:EUR>=0,qty:int>=1,total:EUR=cost * qty
fuel,84.20,1,
dinner for three,`=96.75 / 3`,3,

what Cells reads

	item	cost	qty	total
1	fuel	84.20	1	84.20
2	dinner for three	32.25	3	96.75

every cell holds its type

Files that know each other. A folder of them is a database.

A column typed by another file's key must name a row that exists there, so a typo is caught the moment it is written.

try Fix fodo, or add it to accounts.

---
import accounts from "./accounts.cell"
---
date:date,amount:EUR>0,from:accounts.account,to:accounts.account,vat:EUR=amount * 21 / 121
2026-11-01,950.00,bank,rent,
2026-11-03,42.50,bank,fodo,fodo"fodo" is no key of accounts
2026-11-04,2400.00,salary,bank,

what Cells readsmoney.cell with its imports

	date	amount	from	to	vat
1	2026-11-01	950.00	bank	rent	164.88
2	2026-11-03	42.50	bank	fodo	7.38
3	2026-11-04	2400.00	salary	bank	416.53

row 1, to: "fodo" is no key of accounts

Data that fetches itself. A script fills a table, and the file keeps what it got.

Opening a file never runs anything: you see the rows its last run stored, and the run is yours to start.

rates.cell

---
eurusd = run every day at 18:00 in Europe/Madrid
```ts
export default async function ({ last }) {
  const res = await fetch(`https://api.exchange.example/usdeur?since=${last.date}`)
  return (await res.json()).map(d => ({ date: d.day, rate: String(d.close) }))
}
```
```cells
date:date!,rate:num>0,run:int?
2026-11-03,0.9287,9
2026-11-04,0.9312,9
2026-11-05,0.9306,12
```
```cells runs
run:int!,at:datetime,outcome:ok|failed,error:text?
9,2026-11-04T17:00:03Z,ok,
12,2026-11-05T17:00:04Z,ok,
```
---
date:date,usd:USD>0,note:text?,eur:EUR=usd * eurusd[date].rate
2026-11-03,180.00,two nights,
2026-11-05,42.50,dinner,

eurusd, as storedlast run 12

	date	rate	run
1	2026-11-03	0.9287	9
2	2026-11-04	0.9312	9
3	2026-11-05	0.9306	12

its runsrecorded

	run	at	outcome	error
1	9	2026-11-04T17:00:03Z	ok
2	12	2026-11-05T17:00:04Z	ok

rates.cell, readeur from each day's rate

	date	usd	note	eur
1	2026-11-03	180.00	two nights	167.17
2	2026-11-05	42.50	dinner	39.55

These runs are recorded: this page runs no script. cells run rates.cell runs it, with Bun, when you ask.

Ask the whole folder. Typed tables answer questions.

Every file is a table whose columns are already numbers, dates and references, so a question is a query, never a parse.

---
import accounts from "../accounts.cell"
---
date:date,amount:EUR>0,from:accounts.account,to:accounts.account
2026-11-01,950.00,bank,rent
2026-11-12,64.30,bank,food

every money file, by where it wentdirection: cells sql

$ cells sql . "select a.kind, sum(m.amount) from money m
    join accounts a on m.to = a.account group by a.kind"

	kind	total
1	expenses	1236.80
2	assets	2400.00

Answered here from the engine's typed columns of home/money.cell and travel/money.cell. cells sql, DuckDB over the same columns, is what comes next.

AND NOT

Why not what exists. Each is good at something else.

CSV with a JSON schema beside it: two files that drift: the schema says one thing, the file another, and nothing notices; the types are the header: one file, read and checked as one
Parquet: typed, fast, and a binary: no keys, no references, nothing you can read, diff in git or fix by hand; text you read and diff, keys and references checked; Parquet stays a good export
Oils' TSV8: types in the file for shell pipes; the spec is syntax only, so int, float or decimal is each reader's guess; what a value means is the spec, money exact at its cents, the same in every reader
a spreadsheet: formulas point at B7, which moves when a row is inserted; macros run when a file is opened; formulas name columns and keys; opening a file runs nothing

WHERE IT STANDS

Built

The spec, and a conformance suite every reader runs.
The engine in Rust, passing all of it natively and as WebAssembly: what runs on this page.
Formulas in headers and cells, tables written inline, imports pinned by hash, references checked across files.
cells check, show, pin, and run with Bun.
The table you saw, on the web and on Apple platforms.

Being built, then next

cells lsp, being built now: the problems you saw on this page, in your editor, with hover, completion, go to and rename. Helix first.
A daemon that keeps the schedules and serves every app.
cells sql, DuckDB over the typed columns of a folder.
Charts from the types, with no setup.